Improving Radar Quantitative Precipitation Estimates with Crowdsourced Rainfall Data

Accurate rainfall measurements are of utmost importance for the design of infrastructure, flood warning systems or weather forecasts, among many other applications. While data from weather radars can provide high-resolution information on the spatial variability of precipitation, the derived quantitative precipitation estimates (QPE) are subject to various error sources and therefore uncertain. Rain gauges, on the other hand, are reliable ground-truth observations which lack spatial resolution. National meteorological services have worked on radar technology itself, as well as on processing and gauge-adjustment techniques, to provide high-accuracy QPE products. With the start of the Operational Program on the Exchange of Weather Radar Information (OPERA), researchers can now access radar data on a continental scale. Combined with advances in AI technology and the rising number of personal weather stations (PWS), manifold opportunities for research aiming at improving radar-QPE arise. The project EuRadCA (Improving Rainfall Estimation Across Europe Through Advancements in Radar Data Processing, Use of Crowdsourced Data, and AI) tackles these issues in three separate working groups. Our work at Uni Stuttgart will focus on incorporating crowdsourced data from PWS. The tasks can be subdivided into three work packages aiming at:

1. Improved PWS quality control (QC) using radar data with a focus on extremes
   The first task is the automation, improvement, and development of new quality control methods, such as enhancing an existing bias-correction approach which uses a theoretical gamma distribution. The main goal will be to develop a radar-based QC that assesses the plausibility of both dry stations and high PWS precipitation values, while accounting for a space-time tolerance between radar and gauge measurements.
   
   
2. Intra-pixel variability and conditional radar-gauge uncertainties
   A second step is the quantification of subpixel variability by estimating intensity-dependent correlation structures, which will be used for conditional simulations, constrained by the PWS data, including their uncertainty and the rainfall intensity in the corresponding radar volume. Because this analysis will be undertaken in the radar's spherical coordinates, we will learn how uncertainty increases along the radar beam.
   
   
3. Account for radar-gauge uncertainty during adjustment
   Lastly, we will quantify uncertainty for individual rain gauges and radar pixel-rain gauge pairs. These, together with uncertainty from the individual processing steps in the radar and gauge workflows, will be used to produce confidence intervals for each pixel in the QPE.